v1.0.4

Release Notes - v1.0.4

Highlights

• Added robust stale-lock self-healing to avoid manual unlocks after interrupted runners.
• Reduced default advisory lock wait timeout from 300s to 180s.
• Added integration coverage for double-create runs to catch lock-leak/deadlock regressions.

What Changed

• Lock acquisition hardening:
  • Added lock-holder app name: postgres-branching-lock-holder.
  • Added lock-holder keepalive tuning for faster dead-session detection:
    • LOCK_TCP_KEEPALIVES_IDLE_SEC (default 30)
    • LOCK_TCP_KEEPALIVES_INTERVAL_SEC (default 10)
    • LOCK_TCP_KEEPALIVES_COUNT (default 3)
  • Added stale-lock cleanup on timeout with one retry:
    • LOCK_STALE_AFTER_SEC (default 1800)
• Reduced default lock wait timeout:
  • LOCK_WAIT_TIMEOUT_SEC default changed from 300 to 180.
• Expanded docs:
  • Added lock behavior details and lock tuning environment variables.
• New integration safety test:
  • create run twice in sequence on the same branch.
  • Verifies second run is no-op and no lingering lock-holder query remains.

Validation

• Unit tests:
  • tests/test_create_action.sh
  • tests/test_delete_action.sh
  • tests/test_reset_action.sh
• Integration script syntax:
  • tests/test_create_action_integration.sh

Compatibility

• No breaking input/output changes.
• Existing workflows remain compatible.

Full Changelog: v1.0.3...v1.0.4